docs: frame drift tracer as a workaround

Author: TimesAndPlaces

docs-website/docs/development/tracing.mdx

@@ -358,11 +358,11 @@ Here’s what the resulting log would look like when a pipeline is run:
 <ClickableImage src="/img/55c3d5c84282d726c95fb3350ec36be49a354edca8a6164f5dffdab7121cec58-image_2.png" alt="Console output showing Haystack pipeline execution with DEBUG level tracing logs including component names, types, and input/output specifications" />
 
 
-## Behavioral Drift Monitoring with a Custom Tracer
+## Cross-Run Behavioral Drift Monitoring with a Custom Tracer
 
-Haystack's `Tracer` interface can be used for more than routing spans to a backend — it can also detect **behavioral drift** across pipeline runs. This is useful when your pipeline uses a retrieval-augmented or context-compression component and you want to know whether the agent's effective vocabulary is shifting between sessions.
+Haystack's `Tracer` interface can be used for more than routing spans to a backend — it can also detect **behavioral drift** across pipeline runs. This is a workaround for users who want extra observability today without adding a first-class context-compaction event to Haystack itself.
 
-The example below implements a lightweight `DriftMonitorTracer` that tracks which domain-specific terms (the "ghost lexicon") appear in the first pipeline run but disappear in later runs. A Ghost Consistency Score (GCS) below 0.40 typically signals that the pipeline is losing context-critical vocabulary.
+The example below implements a lightweight `DriftMonitorTracer` that tracks which domain-specific terms (the "ghost lexicon") appear in the first pipeline run but disappear in later runs. A Ghost Consistency Score (GCS) below 0.40 typically signals that the pipeline is losing context-critical vocabulary across runs.
 
 ```python
 import contextlib
@@ -521,5 +521,5 @@ if report["alert"]:
 This pattern requires no changes to Haystack internals. The `Tracer` interface is the only extension point needed. For production use, extend `_on_span_finished` to maintain a per-run rolling window and compare against a configurable baseline depth rather than only the first run.
 
 :::note
-This addresses the behavioral-drift monitoring use case from [#10971](https://github.com/deepset-ai/haystack/issues/10971) using the existing `Tracer` interface — no new hooks required.
+This addresses part of the behavioral-drift monitoring use case from [#10971](https://github.com/deepset-ai/haystack/issues/10971) using the existing `Tracer` interface. It is not a first-class boundary event for summarizers or truncators; it is an explicit workaround based on traced inputs and outputs.
 :::